Advice Needed

[Piper]

For a while now, I've been weekly capturing (NTSC) my favourite 2 hr tv show and encoding it to DVD. I use this method repeatedly to tweak my capture/avisynth skills and I find it's quite effective since I am probably my own best critic (being pickier than most).

What I'm finding now is, friends and family are coming to me asking me to transfer their old VHS tapes to DVD and while I'm comfortable with my own settings for my own viewing, I'm a little hesitant when it comes to capturing for archival purposes.

For example, my avisynth filter string may look something like for direct tv caps (using VBLE codec and capturing at 720x480):

KernelDeint(order=1,threshold=0,sharp=false)
Convolution3D(preset="movieHQ")
BiCublinResize(352,480)

For direct tv caps, I'm pleased with the output and with deinterlacing and a little filtering it makes transition to DivX very easy should I wish to keep that particular show on CD-R.

For VHS (pure interlaced Video) captures I would use HuffYUV and add in some cnr, crop & letterbox to remove garbage around the picture. Depending on the quality of the tape, I may also use some more chroma filtering and reduce it to 352x480. I use HuffYUV in this case as it makes importing into Vegas Video easier (RGB).

While I'm still experimenting with denoisers, my real question here is whether or not to deinterlace. I realise there's no general answer for this, I'm just wondering what others are doing in cases similar to this?

What I normally ask people is "Do you want to be able to watch the dvd on their PC or only on your set top box?". Also, do they want me to try to clean the image, or to simply archive what's on the tape?

My personal view is KernelDeint is good enough that there's no real perceptive loss of quality, esp when viewed back on a typical tv. Even when pouring over it on my monitor several times zoomed in, picking over individual frames can't really notice any real issues.

But is this "best practice"?

[Si]

Since KernelDeint has only been around for a few weeks and we've been capturing for years - it might be a little premature to say using it is "best practice" - but by golly it looks like the holy grail so far :)

Deinterlacing is necessary/needed if your final codec only works well with progressive material and/or you can't afford the bit-rate increase that comes with encoding both fields separately and/or you want to watch it on a PC screen.

So "normally" you'd deinterlace if using an MPEG-4 codec onto a CD.

You might deinterlace/might not if encoding an SVCD (not enough bits to go around sometimes)

You wouldn't deinterlace if encoding high bitrate DVD's for watching on a TV via a standalone DVD player (plenty of bits available -why destroy the information) which is probably what friends and family would do with a DVD.

How much processing is not a question that cannot be answered - trial and error are needed. :)

One more thing - always capture interlaced material with Huffyuv - VBLE (AFAIK) is a YV12 codec and you can end up throwing away chrominance info unneccessarily (you might be capturing duplicate info with Huffyuv but better to err on cautious side)

regards
Simon

[Piper]

Since KernelDeint has only been around for a few weeks and we've been capturing for years - it might be a little premature to say using it is "best practice" - but by golly it looks like the holy grail so far

I agree completely.

You wouldn't deinterlace if encoding high bitrate DVD's for watching on a TV via a standalone DVD player (plenty of bits available -why destroy the information) which is probably what friends and family would do with a DVD.

That's the real question, how much information is being destroyed, even with a high quality (IMHO) filter like KernelDeint?

If I don't deinterlace do I risk more from the inevitable question like: "Why are there lines on my screen when I play your DVD on my computer but not on my tv? What's wrong with your DVD?"

How much processing is not a question that cannot be answered - trial and error are needed.

Agreed. Personally, I prefer noise to over filtering. I'm quite sensitive to this and rarely use anything stronger than C3D's movieHQ preset. If nothing else, it looks a little more true to the original source.

One more thing - always capture interlaced material with Huffyuv - VBLE (AFAIK) is a YV12 codec and you can end up throwing away chrominance info unneccessarily (you might be capturing duplicate info with Huffyuv but better to err on cautious side)

I agree, all my captures done for others have all been done using Huffyuv. I only use VBLE for my own purposes as it uses less space, and requires no colorspace conversions (YV12) -> TMPGEnc. I also prefer BiCublinResize which also works well with a YV12 source. But like I said, only for my own captures as I'm still really only experimenting with this.

[Si]

I only use VBLE for my own purposes as it uses less space, and requires no colorspace conversions (YV12) -> TMPGEnc

If capturing from analog TV there is normally different chroma information on each line - so capturing in YV12 means you are throwing away info at the begining of your process chain.

I haven't done the maths but my intuition says diffuse it at the end :)

regards
Simon